KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Geospatial binary mask
Binary representation where pixels belonging to a specific class are marked as 1 and all others as 0, used to isolate objects or regions of interest in geospatial data.
Geospatial UNet architecture
Variant of the UNet network adapted for satellite image processing, incorporating skip connections preserving spatial details and an encoder-decoder structure for precise segmentation.
Spatially explicit convolutional network
CNN architecture that explicitly integrates spatial relationships and geographic coordinates in its convolution operations to better capture spatial structures in geospatial images.
Normalized spectral indices
Mathematical combinations of spectral bands in satellite images (like NDVI, NDWI) used as additional input channels to improve semantic segmentation of land features.
Spatial multi-resolution pyramid
Processing technique where the same geospatial scene is analyzed at multiple scales simultaneously to capture both fine details and broad contexts in segmentation.
Semantic annotation of tiles
Process of dividing large geospatial images into smaller overlapping tiles, semantically annotated for training deep learning models on extended areas.
Geospatial data augmentation
Specific techniques for transforming satellite images (geometric rotations, lighting variations, simulated atmospheric disturbances) to increase the diversity of the training dataset.
Fuzzy class boundary
Phenomenon where transitions between different land cover classes are gradual rather than sharp, requiring probabilistic or multi-label segmentation approaches.
Hierarchical spatial context
Modeling relationships between objects at different spatial scales (from pixel to region) to improve the semantic coherence of predictions in geospatial segmentation.
Land cover mapping
Main application of geospatial semantic segmentation aiming to produce detailed thematic maps of land use and land cover from satellite imagery.
Temporal change detection
Extension of semantic segmentation where differences between successive segmentations of the same area are analyzed to identify land cover changes over time.
Geographic positional encoding
Technique integrating absolute or relative geographic coordinates into the model's features to capture spatial dependencies and geographic patterns in segmentation.
Spatial feature pyramid
Hierarchical structure of features extracted at different spatial resolutions, combined to produce robust segmentation predictions at multiple observation scales.
Multi-scale spatial attention
Attention mechanism adapted to geospatial data that dynamically weights the importance of regions at different scales to improve the accuracy of semantic segmentation.
Morphological post-processing
Refinement step using morphological operations (erosion, dilation, closing) to eliminate noise and smooth boundaries in geospatial segmentation maps.